Analysis

This comes from the file content/analysis.Rmd.

We describe here our detailed data analysis.

Analysis Motivation

The goal of this data analysis is to explore the relationship between asthma hospitalizations, as a measure of the human impact of air pollution, and the amount of green space in different California counties. We hope to learn whether higher levels of asthma hospitalization correlate to lower proportions of green space.

Although asthma hospitalization rates are not a perfect measure of air pollution, they are strongly linked, and California county data on asthma hospitalization rates is publicly available. Within asthma hospitalization rates, we’ll look specifically at age groups and race to assess the varied hospitalization rates of different groups. Racial makeup of a county is often correlated with socioeconomic status, and therefore we hope to examine whether areas with a greater POC (people of color) population have both higher hospitalization rates and less green space, which are both factors that can correlate with areas of lower socioeconomic status. However, it’s important to note that race is certainly not an exact metric for socio-economic status. Our observations may have implications on how socioeconomic levels of a county are linked to differing levels of asthma hospitalizations or green space, but we will not be conclusively defining that correlation within this study.

To assess each county’s green space, the proportion of county area which is parkland and the number of parks per county are used. Both are included because they look at green space in two different ways, and considering area helps to account for the fact that parks can be drastically different sizes. Park land data is an imperfect measure of tree or plant coverage because urban parks can contain few plants. Rural regions can have large areas of plant coverage remaining in private hands, and these areas can improve air quality despite not being open to the public. However, parks data is publicly available, and does generally give a good idea of the number of parks most public citizens in a county should have access to, as well as the area that these public green spaces cover. It would be helpful to have additional data that analyzes all land cover and divides it into percentages of grass cover, forest cover, building cover, and street cover, or similar categories, but that data was not available at this point.

Some of the major questions we are interested in answering include: Does a higher amount of open green spaces or a higher number of public parks correlate to lower asthma hospitalization rates across California counties? What are the differences in racial and age makeup of hospitalizations across counties? What is the relationship between the racial makeup of asthma hospitalizations and the amount of greenspace in a county? What is the relationship between age makeup of hospitalizations and greenspace? Is there a difference in the relationship between number of open parks and asthma hospitalization rate, and proportion of park land and asthma hospitalization rate? Which serves as a better predictor?

While exploring the data, we first examined how open greenspace and number of parks differed across California counties. We focused on the variables open park land, number of parks, and proportion of park-to-total county land. We discovered that although most counties only use about 10% of their land for parks, counties often have a high number of outlier areas. These areas are census tracts, which contain between 1200-8000 people each, which means some specific areas within each county have much larger amounts of open green space than the average. We also looked generally at the relationship between open park land and age-adjusted hospitalization rate by county but did not see any major correlation.

Modeling and Inference

## [1] "asthmaCA_kids"                 "asthmaCA_kids_2"              
## [3] "asthmaCA_kids_v_adults"        "asthmaCA_race_ethnicity"      
## [5] "clean_parks_data"              "parks_asthmaCA_kids"          
## [7] "parks_asthmaCA_kids_2"         "parks_asthmaCA_kids_v_adults" 
## [9] "parks_asthmaCA_race_ethnicity"

STATS

• creates dataset with the summary statistics by county added (sums and averages for each county)

parks_asthmaCA_kids_v_adults_countystats <- parks_asthmaCA_kids_v_adults %>% group_by(county_name) %>% 
mutate(county_open_park_area = sum(total_open_park_area_sqmiles), county_open_parks = sum(open_parks_tract), county_number_hospitalizations = sum(number_hospitalizations), county_avg_open_park_area = mean(total_open_park_area_sqmiles), county_avg_open_parks = mean(open_parks_tract), county_avg_hospitalizations = mean(number_hospitalizations))

DATA: parks_asthma_CA_kids_v_adults

• Overall Plot of County Tract Open Park Area by County and colored by County Tract Number of Hospitalizations

parks_asthmaCA_kids_v_adults %>% 
group_by(county_name) %>% 
ggplot(aes(x = county_name %>% reorder(total_open_park_area_sqmiles), y = total_open_park_area_sqmiles, color = number_hospitalizations)) + geom_point() + theme(axis.text.x = element_text(angle = 90)) + labs(x = "County Name", y = "County Tract Open Park Area (square miles)", color = "County Tract Number of Hospitalizations") + scale_color_viridis_c(option = "plasma")

• Overall Plot of County Open Park Area by County and colored by County’s Number of Hospitalizations (fewer number of points since there’s one summary point per county, next trying with averages instead of sums)

parks_asthmaCA_kids_v_adults_countystats %>% 
ggplot(aes(x = county_name %>% reorder(county_open_park_area), y = county_open_park_area, color = county_number_hospitalizations)) + geom_point() + theme(axis.text.x = element_text(angle = 90)) + labs(x = "County Name", y = "County Open Park Area (square miles)", color = "County Number of Hospitalizations") + scale_color_viridis_c(option = "plasma")

• Overall Plot of County Average (by tract) Open Park Area by County and colored by County’s Average (by tract) Hospitalizations

parks_asthmaCA_kids_v_adults_countystats %>% 
ggplot(aes(x = county_name %>% reorder(county_avg_open_park_area), y = county_avg_open_park_area, color = county_avg_hospitalizations)) + geom_point() + theme(axis.text.x = element_text(angle = 90)) + labs(x = "County Name", y = "County Average (by tract) Open Park Area (square miles)", color = "County Average (by tract) Hospitalizations") + scale_color_viridis_c(option = "plasma")

• Faceted by Kids v. Adult: Age-Adjusted Hospitalization Rate by County and colored by County Open Park Area

parks_asthmaCA_kids_v_adults_countystats %>% 
filter(!is.na(age_adjusted_hospitalization_rate)) %>%
group_by(county_name) %>% 
ggplot(aes(x = county_name %>% reorder(age_adjusted_hospitalization_rate), y = age_adjusted_hospitalization_rate, color = county_open_park_area)) + geom_point() + facet_wrap(~ strata_name) + theme(axis.text.x = element_text(angle = 90)) + labs(x = "County", y = "Age-Adjusted Hospitalization Rate", color = "County Open Park Area") + scale_color_viridis_c()

• Similar as above: faceted by Kids v. Adult: Number of Hospitalizations by County and colored by County Open Park Area

parks_asthmaCA_kids_v_adults_countystats %>% filter(!is.na(age_adjusted_hospitalization_rate)) %>% group_by(county_name) %>% 
  ggplot(aes(x = county_name %>% reorder(number_hospitalizations),
             y = number_hospitalizations, color = county_open_park_area)) +
    geom_point() + facet_wrap(~ strata_name) + 
  theme(axis.text.x = element_text(angle = 90)) + 
  labs(x = "County", y = "Number of Hospitalizations", color = "County Open Park Area") + scale_color_viridis_c()

• Some models

parks_asthmaCA_kids_v_adults_countystats <- parks_asthmaCA_kids_v_adults_countystats %>%
  filter(!is.na(county_number_hospitalizations), !is.na(county_open_park_area))

mod1 <- lm(county_number_hospitalizations ~ county_open_park_area, data = parks_asthmaCA_kids_v_adults_countystats)
beta <- coef(mod1)
parks_asthmaCA_kids_v_adults_countystats %>% ggplot(aes(x = county_open_park_area, y = county_number_hospitalizations)) + geom_point() + geom_abline(intercept = beta[1], slope = beta[2], color = "red")

DATA: parks_asthmaCA_race_ethnicity

• getting the parks data for each individual census tract

parks_data_by_tract <- parks_asthmaCA_race_ethnicity %>%
group_by(tractcode) %>%
summarize(open_parks_tract = mean(open_parks_tract), tract_area_sqmiles = mean(tract_area_sqmiles), total_open_park_area_sqmiles = mean(total_open_park_area_sqmiles), county_name = county_name)

## `summarise()` has grouped output by 'tractcode'. You can override using the
## `.groups` argument.

• summing to get the total areas for parks and general land for each county

parks_data_by_county <- parks_data_by_tract %>%
group_by(county_name) %>%
summarize(total_county_area_sqm = sum(tract_area_sqmiles), total_county_park_area_sqm = sum(total_open_park_area_sqmiles), county_num_parks = sum(open_parks_tract))

• Creating individualized datasets for each race and white vs. POC

hospitalizations_white <- parks_asthmaCA_race_ethnicity %>%
filter(race_ethnicity == "White") %>%
group_by(county_name) %>%
summarize(number_hospitalizations_white = mean(number_hospitalizations))

hospitalizations_black <- parks_asthmaCA_race_ethnicity %>%
filter(race_ethnicity == "Black") %>%
group_by(county_name) %>%
summarize(number_hospitalizations_black = mean(number_hospitalizations))

hospitalizations_hispanic <- parks_asthmaCA_race_ethnicity %>%
filter(race_ethnicity == "Hispanic") %>%
group_by(county_name) %>%
summarize(number_hospitalizations_hispanic = mean(number_hospitalizations))

hospitalizations_asian_pi <- parks_asthmaCA_race_ethnicity %>%
filter(race_ethnicity == "Asian/PI") %>%
group_by(county_name) %>%
summarize(number_hospitalizations_asian_pi = mean(number_hospitalizations))

hospitalizations_ai_an <- parks_asthmaCA_race_ethnicity %>%
filter(race_ethnicity == "AI/AN") %>%
group_by(county_name) %>%
summarize(number_hospitalizations_ai_an = mean(number_hospitalizations))

hospitalizations_poc <- asthmaCA_race_ethnicity %>%
filter(!STRATA_NAME == "White") %>%
group_by(COUNTY) %>%
summarize(number_hospitalizations_poc = sum(NUMBER_OF_HOSPITALIZATIONS))

• Total Number of Hospitalizations per County

hospitalizations_total <- asthmaCA_race_ethnicity %>% group_by(COUNTY) %>% summarise(NUMBER_OF_HOSPITALIZATIONS = sum(NUMBER_OF_HOSPITALIZATIONS))

• Joining the datasets together via shared variable “county_name”

CA_county_asthma_parks <-
full_join(parks_data_by_county, hospitalizations_white, by = c("county_name" = "county_name"))

CA_county_asthma_parks <- CA_county_asthma_parks %>%
full_join(hospitalizations_poc, by = c("county_name" = "COUNTY"))

CA_county_asthma_parks <- CA_county_asthma_parks %>%
full_join(hospitalizations_hispanic, by = c("county_name" = "county_name"))

CA_county_asthma_parks <- CA_county_asthma_parks %>%
full_join(hospitalizations_asian_pi, by = c("county_name" = "county_name"))

# as the AI/AN category values were all 0 or NA, we decided it was not useful to keep analyzing that information

CA_county_asthma_parks <- CA_county_asthma_parks %>%
full_join(hospitalizations_black, by = c("county_name" = "county_name"))

CA_county_asthma_parks <- CA_county_asthma_parks %>%
full_join(hospitalizations_total, by = c("county_name" = "COUNTY"))

DATA: parks_asthma_CA_kids

• sum and averages of the number of total tract area, open park areas, and open park tracts

sums <- parks_asthmaCA_kids %>% group_by(county_name) %>% mutate(total_tract_area = sum(tract_area_sqmiles), total_open_park_area = sum(total_open_park_area_sqmiles), num_open_parks = sum(open_parks_tract))

avgs <- parks_asthmaCA_kids %>% group_by(county_name) %>% mutate(avg_tract_area = mean(tract_area_sqmiles), avg_open_park_area = mean(total_open_park_area_sqmiles), avg_num_open_parks = mean(open_parks_tract))

• relationship between the number of open parks and hospitalizations

ggplotly(sums %>% ggplot(aes(num_open_parks, number_hospitalizations, col = county_name)) + geom_point() + labs(x = "Count of Open Parks", y = "Hospitalizations", col = "County Name"))

• number of open parks and hospitalizations rates

ggplotly(sums %>% ggplot(aes(num_open_parks, age_adjusted_hospitalization_rate, col = county_name)) + geom_point() + labs(x = "Open Parks", y = "Hospitalizations Rate", col = "County Name"))

• number of total open parks area and the number of hospitalizations, colored with the rate of hospitalization

ggplotly(sums %>% ggplot(aes(x = total_open_park_area, y = number_hospitalizations, col = age_adjusted_hospitalization_rate)) + geom_jitter() + labs(x = "Open Parks Area", y = "Hospitalizations", col = "Hospitalization Rate"))

• same plot as above but shows which county each data point is from

ggplotly(sums %>% ggplot(aes(x = total_open_park_area, y = number_hospitalizations, col = county_name)) + geom_jitter() + labs(x = "Open Parks", y = "Hospitalizations", col = "County Name"))

Flaws and Limitations

• Not all parks contain much green space or plant life which would not relate them to air quality
• In rural areas, people may benefit from the improved air quality from privately held land despite not being able to access the land (land with trees/fields)
• LA’s unique geography makes the development of smog via thermal inversion much more likely. This traps air pollution.
• There are many more contributors to air quality and asthma than accessibility to open green spaces
• We don’t currently have census data on total populations in the counties which is a problem especially for the race and ethnicity data since we don’t have a concept of how large or small the number of hospitalizations are in comparison to the total population. However, we are looking to integrate this data imminently

Note on Attribution

If you are directly quoting from a source, please make that clear. You can show quotes using > like this

> To be or not to be.

To be or not to be.

Also, make sure to provide a link or citation to where you are quoting from.

---

Rubric: On this page

you will

• Introduce what motivates your Data Analysis (DA)
  • Which variables and relationships are you most interested in?
  • What questions are you interested in answering?
• Breadth of the DA
  • Make sure that you ask enough initial questions to explore the different variables in your data.
  • i.e. Do you explore more than just one or two variables? Do you explore a few different relationships or many?
• Depth of the DA
  • When you answer one question, usually more questions arise as well.
  • The depth of the DA is about coming up with and exploring the answers to these questions, often iterating the process a few times.
• Modeling and Inference
  • You should also include some kind of formal statistical model and/or inference. This could be a linear regression, logistic regression, hypothesis testing etc.
  • Explain the techniques you used for validating your results.
  • Describe the results of your modelling and make sure to give a sense of the uncertainty in your estimates and conclusions.
• Explain the flaws and limitations of your analysis
  • Are there some assumptions that you needed to make that might not hold? Is there other data that would help to answer your questions? …
• Clarity Figures
  • Are your figures/tables/results easy to read, informative, without problems like overplotting, hard-to-read labels, etc?
  • Each figure should provide a key insight. Too many figures or other data summaries can detract from this.
• Clarity of Explanations
  • Do you introduce why you are doing each analysis?
  • How well do you explain each figure/result?
  • Do you provide interpretations that suggest further analysis or explanations for observed phenomenon?
• Organization and cleanliness.
  • Make sure to remove excessive warnings, use clean easy-to-read code, organize with sections or multiple pages, use bullets, etc.

NOTE: Your Data Analysis can be broken up into multiple pages if that helps with your organization.